Mapping contaminated soils: using remotely-sensed hyperspectral data to predict pH

This study assessed the feasibility of remote mapping and, thus, monitoring of soils contaminated by acid mine drainage. We report on the use of laboratory and airborne spectroscopy to determine pH. Reflectance spectra were obtained for rock and soil samples collected at our test site, the abandoned Brukunga Pyrite Mine in South Australia, using a laboratory-based Analytical Spectral Devices Inc. (ASD) spectroradiometer. A partial least squares (PLS) regression was used to develop a predictive equation for pH based on the reflectance spectra. The validation results indicated that it is possible to generate satisfactory predictions of pH from spectral data, as demonstrated by the ratio of performance to deviation (RPD) of 1.53, a relatively small root mean square error of prediction (RMSE) of 0.91 and an R-2 value of 0.58. Evaluation of the predictive equation indicated that it depends on diagnostic spectral features related to secondary iron minerals resulting from acid mine drainage (AMD). The presence of these minerals was validated independently using X-ray diffraction (XRD). The predictive equation was applied to airborne hyperspectral data, using a form of remote sensing that simultaneously acquires spatially co-registered images in many spectrally contiguous bands (> 50). Hyperspectral data acquired by the HyMap sensor between 1998 and 2001 were used to produce multi-temporal pH maps. Despite the inaccuracy of the global positioning system (GPS), locations of the validation samples and geographical inaccuracy of the airborne imagery, validation of the maps indicated that pH can be generated reliably from airborne hyperspectral data, as indicated by the relatively small RMSE of 0.57 and the R-2 value of 0.72. These maps demonstrate the potential to provide environmental practitioners, including soil scientists, with a spatially comprehensive view of pH related to AMD conditions. This has implications for the application of remotely sensed hyperspectral data for monitoring soil pH related to AMD conditions, especially in the near future when such data will be available from satellite sensors such EnMap (Stuffler et al., 2009). Some examples of their use include a better understanding of the progress of restoration efforts and or to pinpoint areas where future efforts should be concentrated and to evaluate the extent of downstream impacts, including contamination of soil. More generically, soil pH data are routinely required by soil scientists as part of the suite of data for understanding soil characteristics and the ability to provide spatially comprehensive soil pH data, as demonstrated by this study, would be valuable.